There are open-source LLMs you can run on your own computer if you have a powerful GPU. Models like OLMo and Falcon are made by true non-profits and universities, and they reach GPT-3.5 level of capability.
There are also open-weight models that you can run locally and fine-tune to your liking (although these don’t have open-source training data or code). The best of these (Alibaba’s Qwen, Meta’s llama, Mistral, Deepseek, etc.) match and sometimes exceed GPT 4o capabilities.
The issue with that method, as you’ve noted, is that it prevents people with less powerful computers from running local LLMs. There are a few models that would be able to run on an underpowered machine, such as TinyLlama; but most users want a model that can do a plethora of tasks efficiently like ChatGPT can, I daresay. For people who have such hardware limitations, I believe the only option is relying on models that can be accessed online.
For that, I would recommend Mistral’s Mixtral models (https://chat.mistral.ai/) and the surfeit of models available on Poe AI’s platform (https://poe.com/). Particularly, I use Poe for interacting with the surprising diversity of Llama models they have available on the website.
And there are also free, online hosted instances of those same LLMs in a (relatively speaking) privacy-protecting format from DuckDuckGo, for anyone who doesn’t have a powerful GPU :)
Interesting. So they mix the requests between all DDG users before sending them to “underlying model providers”. The providers like OAI and Anthropic will likely log the requests, but mixing is still a big step forward.
My question is what do they do with the open-weight models? Do they also use some external inference provider that may log the requests? Or does DDG control the inference process?
All requests are proxied through DuckDuckGo, and all personalized user metadata is removed. (e.g. IPs, any sort of user/session ID, etc)
They have direct agreements to not train on or store user data, (the training part is specifically relevant to OpenAI & Anthropic) with a requirement they delete all information once no longer necessary (specifically for providing responses) within 30 days.
For the Llama & Mixtral models, they host them on together.ai (an LLM-focused cloud platform) but that has the same data privacy requirements as OpenAI and Anthropic.
Recent chats that are saved for later are stored locally (instead of on their servers) and after 30 conversations, the last chat before that is automatically purged from your device.
Obviously there’s less technical privacy guarantees than a local model, but for when it’s not practical or possible, I’ve found it’s a good option.
Okay that sounds like the best one could get without self-hosting. Shame they don’t have the latest open-weight models, but I’ll try it out nonetheless.
You can check Hugging Face’s website for specific requirements. I will warn you that lot of home machines don’t fit the minimum requirements for a lot of models available there. There is TinyLlama and it can run on most underpowered machines, but its functionalities are very limited and it would lack a lot as an everyday AI Chatbot. You can check my other comment too for other options.
llama is good and I’m looking forward to trying deepseek 3, but the big issue is that those are the frontier open source models while 4o is no longer openai’s best performing model, they just dropped o3 (god they are literally as bad as microsoft at naming) which shows in benchmarks tremendous progress in reasoning
When running llama locally I appreciate the matched capabilities like structured output, but it is objectively significantly worse than openai’s models. I would like to support open source models and use them exclusively but dang it’s hard to give up the results
I suppose one way to start for me would be dropping cursor and copilot in favor of their open source equivalents, but switching my business to use llama is a hard pill to swallow
Stop depending on these proprietary LLMs. Go to !localllama@sh.itjust.works.
There are open-source LLMs you can run on your own computer if you have a powerful GPU. Models like OLMo and Falcon are made by true non-profits and universities, and they reach GPT-3.5 level of capability.
There are also open-weight models that you can run locally and fine-tune to your liking (although these don’t have open-source training data or code). The best of these (Alibaba’s Qwen, Meta’s llama, Mistral, Deepseek, etc.) match and sometimes exceed GPT 4o capabilities.
The issue with that method, as you’ve noted, is that it prevents people with less powerful computers from running local LLMs. There are a few models that would be able to run on an underpowered machine, such as TinyLlama; but most users want a model that can do a plethora of tasks efficiently like ChatGPT can, I daresay. For people who have such hardware limitations, I believe the only option is relying on models that can be accessed online.
For that, I would recommend Mistral’s Mixtral models (https://chat.mistral.ai/) and the surfeit of models available on Poe AI’s platform (https://poe.com/). Particularly, I use Poe for interacting with the surprising diversity of Llama models they have available on the website.
And there are also free, online hosted instances of those same LLMs in a (relatively speaking) privacy-protecting format from DuckDuckGo, for anyone who doesn’t have a powerful GPU :)
Interesting. So they mix the requests between all DDG users before sending them to “underlying model providers”. The providers like OAI and Anthropic will likely log the requests, but mixing is still a big step forward. My question is what do they do with the open-weight models? Do they also use some external inference provider that may log the requests? Or does DDG control the inference process?
All requests are proxied through DuckDuckGo, and all personalized user metadata is removed. (e.g. IPs, any sort of user/session ID, etc)
They have direct agreements to not train on or store user data, (the training part is specifically relevant to OpenAI & Anthropic) with a requirement they delete all information once no longer necessary (specifically for providing responses) within 30 days.
For the Llama & Mixtral models, they host them on together.ai (an LLM-focused cloud platform) but that has the same data privacy requirements as OpenAI and Anthropic.
Recent chats that are saved for later are stored locally (instead of on their servers) and after 30 conversations, the last chat before that is automatically purged from your device.
Obviously there’s less technical privacy guarantees than a local model, but for when it’s not practical or possible, I’ve found it’s a good option.
Okay that sounds like the best one could get without self-hosting. Shame they don’t have the latest open-weight models, but I’ll try it out nonetheless.
What defines powerful? What if you don’t have the necessary hardware?
You can check Hugging Face’s website for specific requirements. I will warn you that lot of home machines don’t fit the minimum requirements for a lot of models available there. There is TinyLlama and it can run on most underpowered machines, but its functionalities are very limited and it would lack a lot as an everyday AI Chatbot. You can check my other comment too for other options.
llama is good and I’m looking forward to trying deepseek 3, but the big issue is that those are the frontier open source models while 4o is no longer openai’s best performing model, they just dropped o3 (god they are literally as bad as microsoft at naming) which shows in benchmarks tremendous progress in reasoning
When running llama locally I appreciate the matched capabilities like structured output, but it is objectively significantly worse than openai’s models. I would like to support open source models and use them exclusively but dang it’s hard to give up the results
I suppose one way to start for me would be dropping cursor and copilot in favor of their open source equivalents, but switching my business to use llama is a hard pill to swallow